Core Viewpoint - The article discusses the recent release of MiniMax's first inference model, MiniMax M1, which is claimed to have context capabilities comparable to the leading model, Gemini 2.5 Pro [2][10]. Group 1: Model Performance - MiniMax M1 has shown competitive performance in various benchmarks, particularly excelling in the MRCR (Multi-Round Co-reference Resolution) task, achieving an accuracy of 62.8%, which is on par with Gemini 2.5 Pro [3][8]. - The model's architecture includes 456 billion parameters with a MoE (Mixture of Experts) structure, allowing it to handle a maximum context length of 1 million words, significantly surpassing DeepSeek-R1's capabilities [10][12]. - The Lightning Attention mechanism used in MiniMax M1 allows for linear growth in time and space complexity with increasing sequence length, making it more efficient than traditional transformers [8][9]. Group 2: Benchmark Comparisons - In the AIME 2024 logic and mathematics tasks, MiniMax M1 performed adequately, with some tasks showing strong results while others were average [3]. - The MRCR task, which tests a model's ability to understand and differentiate between multiple conversation threads, is highlighted as a significant challenge that MiniMax M1 has managed to tackle effectively [6][8]. Group 3: User Experience and Applications - Users have reported impressive experiences with MiniMax M1, including its ability to accurately translate complex documents and maintain context over long interactions [14][22]. - The model's capabilities extend to creative applications, such as generating narrative content and engaging in interactive storytelling, showcasing its versatility [31][33]. Group 4: Future Expectations - There is anticipation for further developments from MiniMax, particularly in video models and other innovative applications, as the company continues to push the boundaries of AI technology [42][46].

今天一定要来给大家安利一个现在很小众的产品。 虽然它不AI，但是在我用完它的那一刻，我觉得我可以立刻抛弃掉我已经写好的选题，可以不睡觉，也要写给大家看。 这个产品叫NoFeed。 而且从今天起，也开始常驻在我的TAB栏底部，成为我手机上最常用的四大金刚之一。 起初是因为今天老时间23:59，差评发了一篇文章。 我是差评的死忠粉，他们每篇文章我都必看的，当时看到 这个标题，我以为差评又发人物故事了，于是抱着吃瓜的心，点进去看了下。 但是结果就吃了这个产品的安利。 我当场就下载然后付了6快钱。 说实话，我真的很久没有见到过，这种用着，让我觉得有点感动人心的产品了。 这个产品的作用其实超级简单，就一个功能。 帮你搜索一些主流平台的任意关键词。 用XX搜XX。 你可能看到这里，会一脸地铁老头看手机的表情，说，这不就是做了一个搜索集成吗，我为什么不打开小红书直接搜就完事了？我为什么要打开你这 个？ 看看这个小红书搜索的案例。 我不知道你有没有看到一个有趣的情况，就是，这个产品，会让你看不到首页，直接进到搜索结果页。 再结合这个产品的名字，NoFeed。 我想，此时的你，应该能明白，它的核心作用到底是什么了。 帮你，专注 ...

人在去机场的路上，有感而发，随便写写。 我昨天下午终于完成了OKKI 2025的新品发布会。 这是我第一次以发布会顾问+主持人的身份，深度参与一场外贸行业ToB产品的发布。 两个月前，我还对外贸两个字没什么特别的理解。在我认知里，它就是义乌的小商品、深圳的耳机工 厂、福建的丑拖鞋、浙江的男装。我只知道那是一个离钱很近的行业。 后来因为机缘巧合，认识了刘世奇，写了一篇他的专访： 专访刘世奇 - 他用AI设计丑拖鞋，一年卖 了3000万。 也因为刘世奇，跟阿里国际站结识，然后小满科技（OKKI产品的公司、也是阿里国际站的子公司） 正好要开AI新品的发布会，想找一个对外贸感兴趣的AI领域的KOL一起来聊，于是我们就正好，一 拍即合。 但当我真正走进去，跟他们聊完，看到数据，听完几个老业务员、年轻创业者、老板们的真实反馈之 后，我突然意识到，这帮子外贸人嗅觉真的很敏锐，他们已经把AI玩出花了。 不只用来写报告，也不止用来画个图。而是上来就用AI谈单、筛客户、做背调、跟单、复盘、质 检、建图谱、盘老客户，甚至是自动出击。 我是真的服了。 所以我决定写下这篇文字，总结8条我这半个月跟OKKI筹备过程中，和一些AI外贸专家 ...

就在刚刚，夸克官宣了他们最新的夸克高考志愿大模型。 虽然他们没咋提Agent这个词，但是我依然觉得，这玩意比Agent还Agent。 我左想右想，也没想到，夸克会在高考结束的这个时间点，发了可能是我觉得目前AI里，最落地最有用最有社会意义的产品。 高考报志愿Agent。 这个东西，对于广大学子来说，有多有用，我相信每个人都有数。 十几年前，我其实就倒在了志愿填报了，虽然我考的也并不咋地，但是其实后面复盘，发现还是有明显更好的机会。 但是2013年，一个小城市的人家，谁知道，高考志愿有那么多弯弯绕绕的啊。 最后也只能去了，一个普通普通的学校，一个我 可能并没有那么喜欢的专业。 大学期间到没有什么特别的感受，但是当大三开始找实习的时候，真正跟全社会竞争的时候，才能感受到，那种被碾压的压力。 就...完全不是一个起点。 海投了一圈简历，得到了北京一个中厂的面试机会，从广东坐了20多个小时的绿皮火车来到北京，带着我做了1个月的作品集，就为了一次面试。 也可能是面试官觉得我认真，可能觉得作为一个实习生我不仅能做设计还有数据思维，也可能是被我坐了20多个小时的火车就为了这一个面试而打动， 一个实习生居然面了5轮，最后总监 ...

Core Viewpoint - The article discusses the launch of the Seedance 1.0 pro video generation model by Huoshan Engine, highlighting its advanced features and capabilities in AI video production, which positions it as a leading product in the market [1][77]. Group 1: Product Features - Seedance 1.0 pro includes multiple innovative models such as the video generation model, which has gained significant attention and popularity [1][79]. - The model allows for multi-angle combinations and seamless scene transitions, enhancing the storytelling aspect of video creation [9][13]. - It demonstrates high-quality motion rendering, accurately depicting physical dynamics and emotional expressions in generated videos [20][30][57]. Group 2: Performance Evaluation - The performance of Seedance 1.0 pro has been tested across various dimensions, including multi-camera setups, motion quality, emotional portrayal, camera movement, physical dynamics, and stylistic consistency [8][10][49][69]. - The model's semantic understanding is noted to be impressive, effectively translating prompts into visual outputs [15][18]. - The evaluation indicates that Seedance 1.0 pro excels in sports motion, emotional expression, and maintaining stylistic consistency, often exceeding expectations [77][78]. Group 3: Market Position - Seedance 1.0 pro is positioned as a top contender in the AI video generation market, with expectations to maintain its leading status for an extended period [77][78]. - The competitive landscape is acknowledged, with other companies also developing similar technologies, indicating a rapidly evolving market [77][78]. - The pricing for enterprise users is set at approximately 3.67 yuan for every 5 seconds of 1080P video, making it accessible for businesses [79].

Core Viewpoint - The article discusses the performance of various AI models in a simulated high school mathematics exam, highlighting unexpected results and the rapid evolution of AI capabilities in understanding and solving mathematical problems [1][21]. Group 1: Testing Methodology - The testing included previously missing models such as Zhiyu Z1, Kimi 1.5, and Wenxin X1, aiming to provide a comprehensive assessment of AI models' mathematical abilities [3][8]. - Specific scoring rules were established, focusing on correctness of results rather than step-by-step solutions, with each question being run through the models three times to determine accuracy [5][6]. - The inclusion of multimodal questions required models to interpret images, which proved challenging for many, with only OpenAI's model performing adequately [10][12]. Group 2: Results and Rankings - The results were surprising, with models like Xunfei Xinghuo and Doubao achieving high scores of 145 points, excelling in most questions except for a specific one [15][16]. - Qwen3 scored 143.3 points, performing well in answer questions but losing points in fill-in-the-blank sections [16]. - Gemini 2.5 Pro ranked fourth with a score of 139.7 points, while other models like Hunyuan T1 and Wenxin X1 tied for fifth place with slightly lower scores [17][18]. Group 3: Observations and Implications - The article notes the rapid advancement of AI, suggesting that within two years, AI models have reached a level comparable to that of excellent students in high school mathematics [21]. - The author expresses a sense of excitement and surprise at the results, indicating a positive outlook on the future capabilities of AI in educational contexts [22].

这两天，很多媒体都在写用AI考高考题的内容。 我本来真的没打算卷这个选题，因为知道大家肯定都会写，都会卷，我也想休息休息，真的就不打算写了。 但是吧，用AI测语文考试还没啥，但是看了一些用AI做数学考试的文章，真的给我看的一脸地铁老头表情包，就，那个测试方法，也特么太扯淡了。 我觉得既然是考试，那就公平公正的去测试？ 当然，你要是玩整活，那就另谈了。 结果最后得出一些不太靠谱的结论，我觉得还是蛮误导大家的。 客观、公平、公正，是我觉得最核心的标准。 所以我觉得，我想按照我的玩法，再严谨一点的测一下大模纯数学能力型高考，给大家看一下，真实客观的评分。 测试试卷为2025年数学全国一卷。 测试规则如下： 1. 不考解答题（因为给我标准答案我也看不懂，不知道咋给分。。） 2. 所有的题目截图全部使用LaTeX编辑器转成LaTeX文本格式，再扔给大模型进行回答。 LaTeX是学术界最广泛使用的数学公式排版语言，能最精确地表达数学符号，我们考的是模型的数学能力，不是考模型的多模态识图能力，比如 DeepSeek根本就没多模态，用的是OCR提取文本，很可能识别错误，所以截图上传不公平，一律转化成LaTeX格式再进行统一测 ...

Core Viewpoint - The article discusses the launch of PixVerse's domestic version, renamed "拍我AI," and reflects on the rapid evolution of AI video technology over the past 500 days, highlighting the company's journey and achievements in the industry [1][25]. Company Development - PixVerse was established in April 2023, and within a short period, it became one of the leading AI video model companies globally, alongside Runway and PIKA, referred to as the "three giants" in the industry [4][8]. - The company gained significant traction with its ability to generate 4K AI videos, distinguishing itself from competitors who struggled with lower resolutions [8][14]. - The internal testing version of PixVerse was launched in October 2023, leading to a surge in popularity due to its innovative features and user-friendly video templates [8][16]. Market Performance - PixVerse achieved remarkable success in international markets, ranking fourth in the US App Store's free overall chart, and topping various categories in countries like Israel, Turkey, and Saudi Arabia [11][13][14]. - The article highlights the company's strategic focus on overseas markets before launching its domestic version, which was delayed due to resource constraints [11][12]. Product Features and Innovations - The article emphasizes the unique selling proposition of PixVerse, which lies in its user-friendly video templates that cater to the needs of ordinary users rather than just professionals [16][18]. - PixVerse has consistently updated its models, with multiple versions released within a year, showcasing its commitment to innovation and staying competitive in the rapidly evolving AI video landscape [18][20]. Future Outlook - The launch of the domestic version "拍我AI" marks a significant milestone for PixVerse, as the company has expanded its team and resources to enhance its offerings [24][25]. - The article concludes with a sense of nostalgia and anticipation for the future, suggesting that the journey of PixVerse and the AI video industry is far from over, with potential for further growth and development [25].

昨天晚上，即梦的最强AI绘图模型图片3.0，又又又更新了。 内测上线了即梦图片3.0的，智能参考，现在，可以垫图了。 MD，这次连设计师的参考图也一键干碎了。。。 我测了整整一夜，现在是凌晨4点21，我还在写这篇文章。 我人真的傻了，我真的不愿意用一些什么很夸张的词语，但是即梦的绘图，每一次，带给我的震撼，都会觉得，我这么多年的设计师生涯，在AI的进化 速度面前，不值一提。 什么样的言语，都无法比拟直接看图来的直接，直接给你们看效果。 一键改表情包的字，什么叫表情包自由，这就是。 这是一张，很好看的北京的字体设计。 而现在，我很喜欢这个字体设计，我想把北京，变成上海。 你只需要把这张图传给即梦，说，变成上海。 我一定要给你们看看细节，北京的字体设计里面，是有天坛地标的，而上海的设计里面，他自己把地标东方明珠也加上了。 真的，就一句话，太离谱了，真的。 做过设计的人都知道，做这种字体，有多复杂，但是现在，你只要一个效果，一键。 还有可以，继续一句话，做成杭州、新疆、成都。 Prompt：把文字改成"宇宙电波" 还有朋友@倒放 做的，把"九"，改成"十"。 打麻将打的不爽了？把发发发换成胡胡胡。 @阿真Irene ...

Core Viewpoint - The article discusses the considerations for choosing between local and cloud-based AI models, emphasizing the importance of computational requirements and privacy needs when selecting hardware for AI applications [5][6][17]. Group 1: AI Model Deployment - Local deployment of AI models is suitable for applications requiring high computational power and privacy, particularly when handling sensitive data [16][17]. - The article outlines the parameters of AI models, indicating that a model with 1 billion parameters requires approximately 4GB of memory for full precision, while half-precision models can reduce this requirement significantly [11][14]. - For local deployment, models with fewer than 14 billion parameters are generally manageable, while larger models may necessitate high-end GPUs like the RTX 4090 or 5090 [14][19]. Group 2: Hardware Recommendations - The article provides recommendations for laptops suitable for AI applications across different price ranges, highlighting models with specific GPU configurations [26][29][31]. - For a budget of around 5000 yuan, the Mechrevo Aurora X with a 5060 GPU is suggested as a high-value option [26]. - In the 6000 yuan range, the HP Shadow Elf 11 with a 5060 GPU is recommended, while the 7000 yuan range includes upgraded versions of the same model [29][31]. Group 3: Privacy and Security - Local deployment is emphasized as a necessity for applications involving sensitive data, such as business secrets or medical information, to prevent data leaks [17][18]. - The article argues that using local models ensures that all computations are performed on the user's hardware, eliminating the risk of data exposure to third-party services [16][17].